Occams the programmes
For each programme: what did we search, what did it cost, what did we find. Rendered from the programmes' Registers and nothing else. A dated artifact, not a live view.
P1 c2706a09716b (34 records) · P2 23cf22aa6586 (37 records) · P3 e92fda14a7f1 (14 records) · chain verified · generated 2026-10-04T21:29:29+00:00 · repository e7cc38d
Author's build: budgets, the falsifier count and the split come from the author's configuration. No money appears on this page and none can.
Programme 1 · register.jsonl · 34 records · head c2706a09716b Closed
What did we search
Regime classifier frozen at 0434836a176a, index-level on SPY, calibrated on the definition partition only (1993-01-29 to 2003-02-28).
No survey: every question of this programme was registered from a Draft.
Questions registered — depth, with alpha
| Question | Tier | Axis | Mechanism | k | Origin | State |
|---|---|---|---|---|---|---|
| Q-003 | mechanism | regime | In the up regime as labelled by classifier 0434836a176a (index-level), breakout entries carry positive EV in net R because participants under-react to continuation within the regime | 4 | registered from a Draft, not from a survey | null |
| Q-004 | mechanism | regime | In the up regime as labelled by classifier 0434836a176a (index-level), breakout entries carry positive EV in net R because participants under-react to continuation within the regime | 2 | registered from a Draft, not from a survey | null |
| Q-005 | mechanism | price_daily | After 3 consecutive lower closes, a liquid US index ETF carries positive EV in net R over the next 3 bars because a streak of declines exhausts short-horizon sellers and the next days correct (reversal after a streak; ADR-0037) | 4 | registered from a Draft, not from a survey | null |
What did it cost
Alpha by axis · author's build
| Axis | Spent | Budget | Remaining |
|---|---|---|---|
| price_daily | 0.0400 | 0.20 | 0.1600 |
| regime | 0.0600 | 0.20 | 0.1400 |
| cross_sectional | 0.0000 | 0.00 | 0.0000 |
| fundamental | 0.0000 | 0.00 | 0.0000 |
| reserve | 0.0000 | 0.10 | 0.1000 |
Spend is replayed from AlphaSpent records: rate × search-space size. The reserve stays sealed.
Observations consumed
| Question | Partition | Span | Names |
|---|---|---|---|
| Q-003 | measurement | 2003-03-01 to 2019-12-21 | QQQ, SPY |
| Q-004 | measurement | 2003-03-01 to 2019-12-21 | DIA, QQQ, SPY |
| Q-005 | measurement | 2003-03-01 to 2019-12-21 | DIA, IWM, QQQ, SPY |
What did we find
Lab falsifier: 3 of 3 null mechanism verdicts — the count at which the lab closes, declared before the first question.
The lab closed: 3 null verdicts (Q-003, Q-004, Q-005). Nothing registers after this record. The conclusion is drafted from the Register in docs/PROGRAMME-CONCLUSION.md, the author's to adopt.
Verdicts — every one beside where it came from
| Question | Outcome | EV net R | Trades / yr | n | Refused by | Origin |
|---|---|---|---|---|---|---|
| Q-003 | null | -0.0466 | 82.8 | 1,391 | beats-null: random entry under the same geometry does as well; floor: EV per trade in net R is below the declared floor; leave-one-out: fewer than three groups, so robustness cannot be shown | registered from a Draft, not from a survey |
| Q-004 | null | -0.0451 | 124.3 | 2,088 | plateau: the winner's neighbourhood is too small to show a plateau; beats-null: random entry under the same geometry does as well; floor: EV per trade in net R is below the declared floor; leave-one-out: the pooled effect is carried by one group | registered from a Draft, not from a survey |
| Q-005 | null | +0.0513 | 53.2 | 893 | floor: EV per trade in net R is below the declared floor; leave-one-out: the pooled effect is carried by one group | registered from a Draft, not from a survey |
Re-scored under the corrected rules — a diagnostic beside each record, never a verdict
Every outcome above stands as it was reached. The rules that reached them were later found looser than they claimed and were corrected, for records made after (ADR-0047). What those rules would have said of each question is in register/diagnostics.jsonl and in the re-score: a question is first recreated exactly from its stamped source, and only then judged again. It moves no falsifier and spends no alpha.
| Question | As recorded | Recreated first | Under the corrected rules | Annotates |
|---|---|---|---|---|
| Q-003 | null | exactly | would be null; refused by beats-null, floor, leave-one-out, beats-always-long | #9 4bec6db4fbde |
| Q-004 | null | exactly | would be null; refused by plateau, beats-null, floor, leave-one-out, beats-always-long | #21 7d6814ef0aa1 |
| Q-005 | null | exactly | would be refused at measurement, underpowered at its own dispersion; of the five checks, refused by beats-null, floor, leave-one-out, beats-always-long | #32 287691f54b73 |
Programme 2 · programme-2.jsonl · 37 records · head 23cf22aa6586 Stopped
What did we search
Universes, each a record with its calendar frozen
| Universe | Names | Rule | Bias, named |
|---|---|---|---|
| index_etfs | 4 | the four most liquid US-listed broad-index ETFs: S&P 500, Nasdaq-100, Dow 30, Russell 2000 — the class the first program | currently-listed only; all four have survived, which is the class's one declared bias (M0.7); index ETFs carry no single-name survivorship |
| sector_etfs | 11 | the eleven Select Sector SPDR ETFs, one per GICS sector, as listed on 2026-09-12 | currently-listed only; the sector set is rules-based and complete, so no selection bias beyond the class's own; XLRE (2015-10) and XLC (2018 |
| dow_30 | 30 | the thirty constituents of the Dow Jones Industrial Average as listed on en.wikipedia.org/wiki/Dow_Jones_Industrial_Aver | survivorship by construction: a current membership list contains only names that survived and prospered to 2026; every historical verdict on |
| sp_100 | 101 | the 101 constituents of the S&P 100 as listed on en.wikipedia.org/wiki/S%26P_100 on 2026-09-12 (page last edited 2026-09 | survivorship by construction: a current membership list contains only names that survived and prospered to 2026; every historical verdict on |
Regime classifier frozen at 0434836a176a, index-level on SPY, calibrated on the definition partition only (1993-01-29 to 2003-02-11).
Surveys — breadth on the definition partition, zero alpha, no verdict
| Grid | Grid sha | Cells | Universes | Results | Seed | |
|---|---|---|---|---|---|---|
| grid-001 | b98ce89d7cd8 | 38,976 | dow_30, index_etfs, sector_etfs, sp_100 | 80f551b4de5c | 20260912 | the survey page |
Questions registered — depth, with alpha
| Question | Tier | Axis | Mechanism | k | Origin | State |
|---|---|---|---|---|---|---|
| Q2-001 | mechanism | regime | After 4 consecutive lower closes, an S&P 100 constituent as listed on 2026-09-12 carries positive EV in net R over the next 5 bars because a streak of declines exhausts short-horizon sellers and the next days correct (reversal after a streak; ADR-0037) — inside the up regime of the classifier frozen on SPY (F11) | 15 | survey cell f2199b5b297d85bb of results 80f551b4de5c · 38,976 cells screened · sp_100 | supported |
| Q2-002 | mechanism | price_daily | After 4 consecutive lower closes, an S&P 100 constituent as listed on 2026-09-12 carries positive EV in net R over the next 20 bars because a streak of declines exhausts short-horizon sellers and the next days correct (reversal after a streak; ADR-0037) | 15 | survey cell f8500a69db544a3c of results 80f551b4de5c · 38,976 cells screened · sp_100 | null |
What did it cost
Alpha by axis · author's build
| Axis | Spent | Budget | Remaining |
|---|---|---|---|
| price_daily | 0.1500 | 0.20 | 0.0500 |
| regime | 0.1500 | 0.20 | 0.0500 |
| cross_sectional | 0.0000 | 0.00 | 0.0000 |
| fundamental | 0.0000 | 0.00 | 0.0000 |
| reserve | 0.0000 | 0.10 | 0.1000 |
Spend is replayed from AlphaSpent records: rate × search-space size. The reserve stays sealed.
Observations consumed
| Question | Partition | Span | Names |
|---|---|---|---|
| Q2-001 | measurement | 2003-02-12 to 2019-12-17 | AAPL, ABBV, ABT, ACN, ADBE, AMAT, AMD, AMGN, AMT, AMZN, AVGO, AXP, BA, BAC, BKNG, BLK, BMY, BNY, BRK-B, C, CAT, CL, CMCSA, COF, COP, COST, CRM, CSCO, CVS, CVX, DE, DHR, DIS, DUK, EMR, FDX, GD, GE, GILD, GM, GOOG, GOOGL, GS, HD, IBM, INTC, INTU, ISRG, JNJ, JPM, KO, LIN, LLY, LMT, LOW, LRCX, MA, MCD, MDLZ, MDT, META, MMM, MO, MRK, MS, MSFT, MU, NEE, NFLX, NKE, NOW, NVDA, ORCL, PEP, PFE, PG, PM, QCOM, RTX, SBUX, SCHW, SO, SPG, T, TMO, TMUS, TSLA, TXN, UBER, UNH, UNP, UPS, USB, V, VZ, WFC, WMT, XOM |
| Q2-002 | measurement | 2003-02-12 to 2019-12-17 | AAPL, ABBV, ABT, ACN, ADBE, AMAT, AMD, AMGN, AMT, AMZN, AVGO, AXP, BA, BAC, BKNG, BLK, BMY, BNY, BRK-B, C, CAT, CL, CMCSA, COF, COP, COST, CRM, CSCO, CVS, CVX, DE, DHR, DIS, DUK, EMR, FDX, GD, GE, GILD, GM, GOOG, GOOGL, GS, HD, IBM, INTC, INTU, ISRG, JNJ, JPM, KO, LIN, LLY, LMT, LOW, LRCX, MA, MCD, MDLZ, MDT, META, MMM, MO, MRK, MS, MSFT, MU, NEE, NFLX, NKE, NOW, NVDA, ORCL, PEP, PFE, PG, PM, QCOM, RTX, SBUX, SCHW, SO, SPG, T, TMO, TMUS, TSLA, TXN, UBER, UNH, UNP, UPS, USB, V, VZ, WFC, WMT, XOM |
What did we find
Lab falsifier: 1 of 3 null mechanism verdicts — the count at which the lab closes, declared before the first question.
The programme stopped: the author (MaverickHQ) stopped it: Every family the survey holds is beyond both axes; the two verdicts, one by each winner rule, have answered what grid-001 could ask at this floor. Nothing registers after this record. The conclusion is written from the Register by python -m occams conclude (M12.8), the author's to adopt.
Verdicts — every one beside where it came from
| Question | Outcome | EV net R | Trades / yr | n | Refused by | Origin |
|---|---|---|---|---|---|---|
| Q2-001 | supported | +0.2134 | 412.7 | 6,946 · 1 missed | none | survey cell f2199b5b297d85bb of results 80f551b4de5c · 38,976 cells screened · sp_100 |
| Q2-002 | null | +0.1089 | 756.5 | 12,732 · 1 missed | floor: EV per trade in net R is below the declared floor | survey cell f8500a69db544a3c of results 80f551b4de5c · 38,976 cells screened · sp_100 |
Re-scored under the corrected rules — a diagnostic beside each record, never a verdict
Every outcome above stands as it was reached. The rules that reached them were later found looser than they claimed and were corrected, for records made after (ADR-0047). What those rules would have said of each question is in register/diagnostics.jsonl and in the re-score: a question is first recreated exactly from its stamped source, and only then judged again. It moves no falsifier and spends no alpha.
| Question | As recorded | Recreated first | Under the corrected rules | Annotates |
|---|---|---|---|---|
| Q2-001 | supported | exactly | would be null; refused by floor, beats-always-long | #19 37327a3559ec |
| Q2-002 | null | exactly | would be null; refused by floor | #33 bb6c9b72b0ee |
Shrinkage from screening
| Question | Survey cell | Screened | EV, definition | EV, measured | Δ EV | Margin, definition | Margin, measured | Δ margin | Verdict |
|---|---|---|---|---|---|---|---|---|---|
| Q2-001 | f2199b5b297d85bb | 38,976 | +0.256 | +0.213 | -0.042 | +0.239 | -0.001 | -0.240 | supported |
| Q2-002 | f8500a69db544a3c | 38,976 | +0.333 | +0.109 | -0.224 | +0.201 | +0.075 | -0.127 | null |
Net R per trade. The definition numbers are the survey cell's (the calibration set, zero alpha); the measured ones are the winner cell's on the measurement partition against always-long at the same geometry. Every cell screened is counted beside every question registered from it (N6).
Programme 3 · programme-3.jsonl · 14 records · head e92fda14a7f1 Stopped
What did we search
Universes, each a record with its calendar frozen
| Universe | Names | Rule | Bias, named |
|---|---|---|---|
| index_etfs | 4 | the four most liquid US-listed broad-index ETFs: S&P 500, Nasdaq-100, Dow 30, Russell 2000 — the class the first program | currently-listed only; all four have survived, which is the class's one declared bias (M0.7); index ETFs carry no single-name survivorship |
| sector_etfs | 11 | the eleven Select Sector SPDR ETFs, one per GICS sector, as listed on 2026-09-12 | currently-listed only; the sector set is rules-based and complete, so no selection bias beyond the class's own; XLRE (2015-10) and XLC (2018 |
| dow_30 | 30 | the thirty constituents of the Dow Jones Industrial Average as listed on en.wikipedia.org/wiki/Dow_Jones_Industrial_Aver | survivorship by construction: a current membership list contains only names that survived and prospered to 2026; every historical verdict on |
| sp_100 | 101 | the 101 constituents of the S&P 100 as listed on en.wikipedia.org/wiki/S%26P_100 on 2026-09-12 (page last edited 2026-09 | survivorship by construction: a current membership list contains only names that survived and prospered to 2026; every historical verdict on |
Regime classifier frozen at 0434836a176a, index-level on SPY, calibrated on the definition partition only (1993-01-29 to 2003-02-11).
Surveys — breadth on the definition partition, zero alpha, no verdict
| Grid | Grid sha | Cells | Universes | Results | Seed | |
|---|---|---|---|---|---|---|
| grid-002 | 1716f9f1c506 | 38,976 | dow_30, index_etfs, sector_etfs, sp_100 | f5b7a3470c77 | 20260920 | the survey page |
Questions registered — depth, with alpha
| Question | Tier | Axis | Mechanism | k | Origin | State |
|---|---|---|---|---|---|---|
| Q3-001 | mechanism | regime | After a close below its 20-bar average while above its 200-bar average, a US sector ETF carries positive EV in net R over the next 10 bars because a dip inside an uptrend is bought back (reversal conditioned on trend; ADR-0037) — inside the up regime of the classifier frozen on SPY (F11) | 15 | survey cell e02384ae41677547 of results f5b7a3470c77 · 38,976 cells screened · sector_etfs | registered, refused at measurement, unresolved |
What did it cost
Alpha by axis · author's build
| Axis | Spent | Budget | Remaining |
|---|---|---|---|
| price_daily | 0.0000 | 0.20 | 0.2000 |
| regime | 0.1500 | 0.20 | 0.0500 |
| cross_sectional | 0.0000 | 0.00 | 0.0000 |
| fundamental | 0.0000 | 0.00 | 0.0000 |
| reserve | 0.0000 | 0.10 | 0.1000 |
Spend is replayed from AlphaSpent records: rate × search-space size. The reserve stays sealed.
What did we find
Lab falsifier: 0 of 3 null mechanism verdicts — the count at which the lab closes, declared before the first question.
The programme stopped: the author (MaverickHQ) stopped it: At the declared numbers no family on grid-002 is registrable: the one question the regime axis could buy, Q3-001, was refused at measurement before a verdict, and what remains on either axis affords no survey family. Whether a margin the screen finds transfers was not answered on this grid at these numbers. Nothing registers after this record. The conclusion is written from the Register by python -m occams conclude (M12.8), the author's to adopt.
No verdict yet.
Re-scored under the corrected rules — a diagnostic beside each record, never a verdict
Every outcome above stands as it was reached. The rules that reached them were later found looser than they claimed and were corrected, for records made after (ADR-0047). What those rules would have said of each question is in register/diagnostics.jsonl and in the re-score: a question is first recreated exactly from its stamped source, and only then judged again. It moves no falsifier and spends no alpha.
| Question | As recorded | Recreated first | Under the corrected rules | Annotates |
|---|---|---|---|---|
| Q3-001 | Refused at measurement · unresolved | no | not re-scored: its refusal stamps no engine | #12 f4fd2b572b24 |